developerjeremylive/s1-mini-etheroi

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.8BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 20, 2026License:otherArchitecture:Transformer Featherless Exclusive Cold

S1-mini by Superwhisper is a 0.6 billion parameter text normalizer, fine-tuned from Qwen3-0.6B, designed to clean raw speech-to-text (ASR) output. It rewrites transcripts by removing fillers, resolving false starts, applying punctuation and capitalization, and rendering spoken numbers, dates, times, currency, and email addresses into written form. This English-only model achieves 94.8% token accuracy on a held-out test set and runs efficiently on laptop CPUs, making it ideal for post-processing ASR output in dictation apps or meeting tools.

Loading preview...

S1-mini: A Specialized Text Normalizer

S1-mini, developed by Superwhisper, is a compact 0.6 billion parameter language model specifically fine-tuned from Qwen3-0.6B to act as a text normalizer for speech-to-text (ASR) output. Unlike general-purpose LLMs, S1-mini performs a single, focused task: transforming raw, unpunctuated ASR transcripts into clean, written English.

Key Capabilities

  • Text Normalization: Rewrites raw ASR output by removing filler words (e.g., "um," "uh"), resolving false starts and self-corrections, and applying correct punctuation and capitalization.
  • Format Conversion: Renders spoken numbers, dates, times, currency, and email addresses into their standard written forms.
  • Contextual Styling: Offers control over output style (casual, semi-casual, semi-formal, formal), structure (prose, lists), and context (general, email) via a specific control line in the input.
  • High Accuracy: Achieves 94.8% token accuracy on a held-out English test set of 7,519 cases.
  • Efficiency: With 0.6B unique parameters (0.8B reported due to tied embeddings), it is designed to run comfortably on laptop CPUs, with GGUF builds available for llama.cpp, Ollama, and LM Studio.

When to Use S1-mini

This model is ideal for applications requiring a post-processing step for ASR output, such as dictation apps, meeting-notes tools, live captioning, or voice-driven editors. It is not a chat model and will not follow general instructions; its strength lies in its specialized, deterministic transformation of speech transcripts into readable text. Users must adhere to the specified system prompt and control line format for optimal performance, and greedy decoding is recommended.

Finetunes
1 models